Software scans your Facebook PHOTOS to work out if you're a hipster, goth or a surfer - and customises the adverts you see
Posted 12/18/2013 12:25:00 PM

Your Facebook posts and the pages you like  already give away a lot about you including what music you listen to, or your  favourite restaurant.

But now researchers are using people’s  Facebook photos to identify what sub-culture they belong to, from goths and  bikers to hipsters and surfers.

They have created an algorithm that can  identify certain elements associated with each of these groups, such as  distinctive tattoos, hats and jewellery, before scanning other photos to find  similar people.

 
 

HOW DOES THE SOFTWARE WORK?

Researchers used a so-called ‘multi-label  classification algorithm’.

The algorithm scans photos that have already  been tagged, and looks for identifying features. For example, if a picture is  tagged as ‘dog’, it will pick out the shapes and sizes of objects in the image.

It will then scan other photos looking for  similar features and elements that match, or closely resemble, the original dog.

The software assumes that the new image  contains similar elements as the original image and tags it with the same label.

This is called a ‘parts and attributes’ approach. 

The University of California researchers used  this model to scan photos of elements most commonly associated with different  sub-cultures. 

It started by scanning the head, neck and  arms for tattoos. It then checked for certain haircuts or jewellery, as well as  the type of clothing and the colours the subjects wore.

These results could then be used by  advertisers and websites to customise the type of content these people see, or  the products they’re more likely to buy.

To do this, researchers from the University  of California used a so-called ‘multi-label classification algorithm’. It is  already used by certain photo editing software, for example, to help tag and  categorise images.

Read more

Posted By: Producer Thomas  
  • 0
Comments are closed.
- advertisement -
- advertisement -
- advertisement -